Consulting Work Blog Contact
AI Integration Consultant · ex-automotive SWE · M.Sc. @ KIT

Alexey Berezka

I integrate AI into real businesses — and build my own in public.

I help companies in Germany and the EU turn "we should use AI" into working systems people actually use — built with the rigor I learned shipping ex-automotive software, so they hold in production, not just in the demo.

10+
Years engineering
200+
AI agents orchestrated
~95%
LLM cost cut
2,000
Docs in RAG engine

AI systems I've built and run in production

Not demos. Each of these runs on my own infrastructure, every day.

🤖
Runs 24/7 in production

JARVIS — Multi-Agent AI Operating System

Problem Generic chatbots don't act — they answer. I wanted an assistant that plans and executes real multi-step tasks across my tools.
Approach A self-hosted system with voice in/out (Whisper), email, calendar and smart-home tools, and a meta-orchestrator that routes work across 200+ specialist agents — autonomously planning, delegating and executing.
Python Claude API Whisper MCP Docker
📈
~95% lower model cost

DeepSignal — Autonomous LLM Pipeline

Problem LLM-driven market analysis is powerful but expensive and fragile — easy to overspend, hard to trust.
Approach An end-to-end pipeline that ingests market & news data, runs LLM analysis and publishes signals on schedule with zero manual touch — a vendor-neutral model adapter plus a shadow-eval production gate, hardened by backtesting and a 54-agent adversarial review.
Python LLMOps Eval gating Telegram API
🔎
2,000 docs · sub-second

Hybrid RAG Knowledge System

Problem Keyword search misses meaning; pure vector search hallucinates context. Neither alone is good enough to ground an LLM.
Approach Retrieval-Augmented Generation over ~2,000 documents combining dense embeddings (bge-m3), sparse FTS5 and a reranker — daemon-served with nightly reindex for fast, grounded, citable answers.
RAG bge-m3 FTS5 Reranker
🏗️
Vendor-portable stack

Self-Hosted AI Infrastructure & MCP

Problem Lock-in to one cloud or one model vendor is a business risk, not just a technical one.
Approach A full production stack on Hetzner — Docker, nginx, systemd, CI/CD (GitHub Actions), FastAPI services and custom Model Context Protocol servers — designed so the LLM backend and data stay portable by default.
Docker nginx systemd CI/CD FastAPI
🌐
30 languages · live

Web Design & Development

Problem Professionals lose clients to sites that look generic and don't rank in their market's language.
Approach Clean, fast multilingual sites — this one runs in 30 languages — built on Hugo with native-reading copy and SEO per market. Recent client: a sworn DE/RU/IT court translator.
Hugo i18n SEO Nginx

🔬 KIT Research — AI Adoption Decision Model

My M.Sc. thesis at KIT: a Multi-Criteria Decision Model that tells manufacturing SMEs where AI image-processing is actually worth adopting — and where it isn't. Built on expert interviews across four industries, quantitative analysis and AHP modeling, it replaces gut-feel AI decisions with a repeatable methodology. Earlier theses: stratospheric-balloon gondola electronics (Ing., UWB Pilsen 2021); multi-scale phase rangefinding for spacecraft tracking (B.Sc., RTU 2018).

KIT HECTOR School MCDM Industry 4.0 AI Adoption SME Research
📋Expert Interviews
🎤Quantitative Analysis
📈MCDM / AHP

From factory floor to AI deployment

AI & Digital Transformation May 2026 — present

Kyocera Automotive and Industrial Solutions

Dietzenbach, Germany

AI and digital transformation work inside a German industrial environment — applying engineering practice and M.Sc. research to real operations.

AI Adoption Digital Transformation Industry 4.0
Executive M.Sc. 2024 — now

Karlsruhe Institute of Technology

KIT HECTOR School

Executive M.Sc. at KIT HECTOR School (90 ECTS, ~20 months) — KIT's technology business school for working engineers. Curriculum interleaves engineering depth (digital platforms, AI, software, security) with management breadth (strategy, decisions, finance, innovation). Specialisation in Autonomous Robotics — smart manufacturing and Industry 4.0. Built to bridge IT and business — engineers leading digital transformation, not just executing it. The programme rebuilt how I think about technology decisions: not "what's technically possible" but "what's the right thing given strategy, capability, and risk."

Strategy Information Systems Decision Theory Digital Transformation
Embedded Software Developer 2019 — 2023

ZF Pilsen

Commercial Vehicles Division — Tier 1 Automotive Supplier

Shipped embedded software for production ECUs in a regulated, safety-critical environment. Cut validation cycles 25% by automating HiL testing pipelines.

C/C++ HiL/SiL Jenkins Git CAN
Engineer 2017 — 2018

Typhoon Research & Production Association

Aviation Weather Systems

Designed measurement systems for aviation weather monitoring. Developed sensor-based data acquisition in a regulated environment requiring documentation rigor.

PCB Design MATLAB Electronics
Data Analytics Engineer 2016 — 2017

Stream

Analytics & Data

Built predictive models for customer segmentation and automated reporting pipelines — eliminating 40% of manual analyst workload.

SAS Python SQL ETL

Technology I work with

AI & Machine Learning
🤖
Agentic AI / Multi-Agent
🔗
RAG / Vector Embeddings
📐
Prompt Engineering
📈
LLMOps / Eval
⚙️
MCP Protocol
🧰
Claude Agent SDK
GPT-5 / Gemini 2.5
🏠
Ollama / Local LLM
🎤
Whisper STT
🛡️
EU AI Act / GDPR
📊
MCDM / AHP
Engineering & Development
🐍
Python
🔧
C / C++
🗄️
SQL
📡
CAN / Embedded
📈
SAS / ETL
🔬
MATLAB
Infrastructure & DevOps
🐳
Docker
🌐
Nginx
🚀
Hetzner VPS
🔀
Git / CI-CD
n8n / Automation
📦
Telegram API

Most AI work stops at the slide deck. I integrate it into the business — and stay until it holds.

I'm Alexey — an AI integration consultant. I put AI into real businesses and build my own tools in public: a personal assistant that wires together Claude, OpenAI and Perplexity, ML research on trading signals, and this site you're reading. I design them, ship them to my own servers, and run them every day.

At KIT I'm finishing an M.Sc. on how manufacturers actually adopt AI — my thesis looks at AI-based image processing in manufacturing SMEs, grounded in expert interviews and real decision criteria rather than vendor hype. It keeps the work I do for clients honest about what AI can and can't deliver.

Before AI I spent a decade in software engineering for automotive and aviation — automotive ECUs at ZF, aviation measurement systems — where nothing ships until it holds under real conditions. That's the bar I bring now: AI that survives production, not just the demo. Off-screen you'll find me in the gym, on the road, or running the next experiment — in four languages (EN · RU · DE · CZ).

0
Years exp.
0
Lang. spoken
0
Industries
AB
Based in Karlsruhe, Germany
Education B.Sc. RTU · Ing. UWB Plzeň · M.Sc. KIT (in progress)
Research How manufacturers adopt AI (M.Sc. thesis)
Languages RU · EN · DE · CZ
Status AI integration consultant · building in public · open to roles & consulting

What keeps me sharp

🏋️

Gym & discipline — training most days.

📷

Photography — I shoot and share on Instagram.

🖨️

3D printing & design — functional prints and STL models I publish on MakerWorld.

🌍

Languages — Russian, English, German and Czech; Italian next.

I write down what I build, most nights

Vibes per Token is my build-in-public journal on AI, agents and automation. What shipped, what broke, what it cost. English and Russian on Telegram.

See the journal

An engineer who speaks both languages

I sit between the server room and the boardroom. I can write the code, secure the infrastructure and design the data pipeline — and then explain to leadership, in plain business terms, what it costs, what it returns and where the risk is. AI built on EU data boundaries, with GDPR and the EU AI Act treated as design constraints, not afterthoughts.

Builds in production
Everything on this page runs on my own servers, daily — not in a demo sandbox.
Responsible & EU-ready
GDPR, EU AI Act and data-sovereignty considered from day one, not bolted on.
Vendor-neutral
Portable stacks — your data and logic never get locked to one model or cloud.

Find your starting point

For CEOs & founders

Turn "we should use AI" into a prioritized, costed roadmap with clear ROI — and a build partner who delivers it.

Scope a roadmap
For CTOs & engineering leads

Add senior hands for agentic AI, RAG, LLMOps and self-hosted infrastructure — without the year-long hiring cycle.

Talk architecture
For recruiters

Hiring for AI & digital transformation in Germany? I'm available from October 2026, on-site in the Rhine-Neckar / Frankfurt area or remote across the EU.

See my CV

Life & Work on Instagram

AB
@alexey.berezka
AI builder · 3D printing · Life in Karlsruhe 🇩🇪
Follow on Instagram

Let's put AI to work in your organization

Whether you're scoping an AI strategy, need someone who can actually build it, or are hiring for AI & digital transformation — let's talk. Based in Germany, available from October 2026, remote across the EU.

Open to AI roles & AI-integration consulting · Germany